Our team was refactoring the codebase to follow ML pipeline best practices, but not enough attention had been paid to problem/task formulation, label quality etc.